Frost & Sullivan Releases '2026 China AI4S Scientific Computing Infrastructure Market Study'"

AI4S Scientific Computing Infrastructure refers to the market and industry cluster that provides products, system construction, and related investments in scientific computing infrastructure for AI4S applications such as universities, research institutions, and scientific platforms. It is not a single computing center but a comprehensive computing foundation composed of intelligent computing clusters, high-speed interconnected networks, high-performance parallel storage, and full-stack software platforms. The scope of this study includes infrastructure products such as computing clusters, AI servers, storage, networks, and liquid cooling, as well as system construction investments.

Looking back at the development history, the industry has gone through four stages: cloudification of general computing power (2015-2018), clustering of scientific computing power (2019-2021), initial stage of ultra-intelligent integration (2022-2023), and platformization of scientific research processes (2024-2025). Starting in 2026, it enters the stage of large-scale construction. As tasks such as training large scientific models, simulating complex systems, automated experiments, and intelligent research agents are gradually implemented on a large scale, AI4S construction is expanding from research groups to university and college levels, regional levels, and national scientific platforms. Market competition focuses on system integration, platform operation, adaptation to disciplinary scenarios, and scientific research ecosystem construction, shifting from hardware delivery.

The Chinese AI4S Scientific Computing Infrastructure Market is moving from pilot construction to large-scale deployment. The report indicates that the potential market size of AI4S infrastructure in Chinese research institutions is expected to grow from 420.7 billion yuan in 2025 to 760.7 billion yuan in 2027, reaching 1,480.1 billion yuan by 2030, with a compound growth rate of approximately 28.6% from 2025 to 2030.

Continuous growth in research investment provides a solid foundation for market expansion. From 2020 to 2024, the total R&D funds of Chinese universities and government-owned research institutions increased from 5,291.3 billion yuan to 7,297.1 billion yuan, with an average annual compound growth rate of about 8.4%. It is expected to reach 10,042.9 billion yuan by 2030. Among them, university R&D funds grew faster, with the proportion expected to increase from 35.6% in 2020 to 47.9% in 2030. At the same time, scientific research is entering the stage of PB-level data, tens of millions of samples, and hundreds of millions of atoms of simulation, driving scientific computing clusters to evolve towards ten-thousand-core parallel and storage-computing coordinated architecture.

In terms of application maturity, AI4S applications in university and research institutions are still in the transition period from pilots to scale: local research groups have the highest proportion at 34%; multiple laboratories and multiple disciplines have expanded applications at 27%; only 15% have formed platform-based applications at the university or college level. From 2025 to 2027, market growth will mainly come from leading university-level computing platforms, research institution AI4S platforms, national key laboratories' intelligent upgrades, and national and regional scientific computing nodes. After 2028, the market will enter a stage of simultaneous new construction and existing upgrade, with needs for node expansion, storage expansion, network upgrade, liquid cooling transformation, scheduling optimization, and adaptation to scientific research software stacks continuing to emerge.

In the past, supercomputing centers primarily handled high-precision deterministic computing tasks such as physical simulation, molecular dynamics, climate modeling, and material calculation, while intelligent computing centers focused on large model training and deep learning inference. These two types of facilities were relatively independent in terms of construction entities, resource scheduling, software environment, and application scenarios. AI4S is breaking this boundary. Scientific research is gradually shifting to a integrated paradigm combining physical models, neural networks, scientific data, and experimental feedback, and infrastructure is evolving from decentralized construction to a unified platform for ultra-intelligent integration. In 2026, the national supercomputing internet core node launched a 100,000-core ultra-intelligent integration cluster Shuguang 8000 (Peak), integrating high-precision scientific computing and AI training/inference capabilities into a unified scientific computing system. Ultra-intelligent integration has entered the stage of large-scale deployment from a technical concept.

On this basis, research institutions are shifting their focus from the number of GPU/NPU units, theoretical peak performance, and node size to multi-core parallel efficiency, network communication, storage throughput, task scheduling, and long-term stable operation capabilities. If network, storage, or scheduling become bottlenecks, a large amount of theoretical computing power will be wasted on communication waiting, data loading, and resource idleness. Evaluation criteria are also moving closer to application effectiveness. Whether it can improve model accuracy, shorten development cycles, reduce trial-and-error costs, and form replicable scientific and industry applications is becoming the core criterion for measuring platform value.

At the application level, AI4S is prioritized in scientific research fields with large data volumes, complex models, long computing cycles, high simulation accuracy requirements, and high trial-and-error costs. In the field of life sciences, protein structure prediction, molecular property prediction, and candidate drug screening are forming a closed loop from computational prediction, candidate screening to experimental verification. In the fields of materials and chemical energy, material databases, machine learning potential functions, and high-throughput simulation are driving research from empirical trial-and-error to model-driven design. In Earth system science, AI weather models are transforming some of the high-cost numerical iteration forecasting capabilities into high-throughput, low-latency inference capabilities, and physical models combined with AI are becoming an important path for high-resolution local weather simulation.

As AI4S moves from pilot construction to platform-based deployment at university, college, regional, and national levels, the core of industry competition is changing. Early competition mainly focused on the number of servers, peak computing power, and cluster size, with the emphasis on building clusters; now, research institutions are more concerned with whether infrastructure can support long-term stable operation of real scientific research tasks. The report indicates that the competitive barriers in the industry are shifting from single hardware performance to systematic capabilities covering customer and project acquisition, effective computing power and architecture, system engineering delivery, basic software and platform productization, scientific research scenarios and applications, and domesticization and supply chain security, progressing layer by layer from hardware, system, platform, to scenarios.

Under this trend, in the short term, manufacturers with the ability to deliver complex systems, platform productization, and basic software adaptation are more likely to obtain large-scale project opportunities. In the long term, manufacturers with benchmark scientific research customers, application co-creation capabilities, and long-term operational services are more likely to establish sustainable competitive advantages.

Currently, the Chinese AI4S Scientific Computing Infrastructure market has formed a competitive landscape involving server and system manufacturers, comprehensive technology companies, operators, and computing power service providers. Manufacturers such as Sunway Technology, Huawei, Inspur Information, Parallel Technology, China Telecom, and China Mobile have formed differentiated advantages based on their respective resource bases: server and system manufacturers focus on large-scale cluster delivery, ultra-intelligent integration architecture, and scientific software adaptation; comprehensive technology companies focus on AI computing power base and software ecosystem; operators and computing power service providers have advantages in regional public computing power, computing power network, and long-term operational services.